grill (idle)
challenges my plan until no loose assumption is left
$ grill-me "agents section v2"
> what breaks if we skip it?
> who maintains it in six months?Explicit boundaries and systems you can explain. No fluff.
Full Stack with product ownership · LLMs and RAG running in production · Mentoring and engineering standards for teams
Real situations I lived through, the problem, the call I made, and why.
context first · spec before code · human review
I use spec-driven development to work with agents. The plan is discussed, stress-tested and written down with its constraints before any code exists. The implementation is checked against that plan, and whatever fails review goes back to planning. I am turning that process into a chain of specialized agents, which I am currently testing on my own projects.
challenges my plan until no loose assumption is left
$ grill-me "agents section v2"
> what breaks if we skip it?
> who maintains it in six months?a planner drafts the plan and a critic tries to break it
$ endev-plan
✓ .plans/home-polish.md
critic · 0 blockersagents build in waves, each one in its own area
$ endev-execute .plans/home-polish.md
✓ wave 1 · web-frontend ×2 · worker
✓ wave 2 · seo-aeointegration tests on the seams between tasks
$ endev-test .plans/home-polish.md
✓ e2e + axe · light and darkscores 0 to 100 against the plan. Below 90 it goes back to the plan and gets fixed
$ endev-review .plans/home-polish.md
! round 1 · 84/100 · 3 findings
✓ round 2 · 93/100validates, shapes the commit history and pushes
$ endev-ship .plans/home-polish.md
✓ tsc · eslint · knip · tests
✓ pushedReal incidents, architecture decisions, opinions on AI and the occasional reflection on how we work. I write them when something breaks, or when something doesn't add up.
Not just today's stack — tools and patterns from years of projects and notes. Not a logo wall: a map of what I've used, still use, and would reach for again. Hover any entry to see the pattern behind it.Not just today's stack — tools and patterns from years of projects and notes. Not a logo wall, a map of what I've used, still use, and would reach for again. Tap + on any entry to see the pattern behind it.
04
Spec-driven development, agent skills, MCP servers — applied AI in production, not just RAG.
Spec-driven development, agent skills, MCP servers — applied AI in production, not just RAG.
Spec-Driven Development (SDD)
Write the spec before the agent writes code — Grill.me interviews the plan until every branch is resolved, then a fresh agent builds against it.
hover to inspecttap + to inspect
# spec.md
## Requirements
## Acceptance criteria
// the agent implements against this, not vibesClaude Code · Cursor
Daily agentic coding — from Cursor's Tab autocomplete to full agent workflows with custom skills.
/anti-cliche # custom skill, runs before every commitMCP Servers
Wiring agents to live data — this site's own content comes from a Sanity MCP server.
claude mcp add Sanity -t http https://mcp.sanity.ioAgent Skills · Multi-agent
Grill.me-style multi-agent flows — one agent interviews the plan, another builds it.
phase-1.md → new conversation → phase-2.md
# context reset between phasesContext Engineering
Splitting large plans into phases before context runs out mid-task.
# plan.md
## Phase 1: schema
## Phase 2: API
## Phase 3: UIRAG · LangChain
Semantic search over internal knowledge.
const docs = await retriever.invoke(question);
return llm.stream({ docs, question });Vector Search
Embeddings + pgvector for discovery.
select 1 - (embedding <=> $1) as score
order by score desc;From freelance frontend to leading product, platform and applied AI. Each step left something I still use.
2026
Senior role with a product focus
I am interested in a role where technical judgment turns into product. I want to build close to the customer, make architecture calls with real context and help the team grow. I have led products and teams before, and I want that experience to reach further.
If you made it this far, you probably want to know a bit more about me. I've spent many years building software, and what matters to me hasn't changed since the beginning, watching an idea turn into something that actually works.
Code is only part of the job. Some days it's product thinking, other days architecture, watching a deploy, or staying a little longer to figure out why something that worked yesterday stopped working today. That's production. It has a personality of its own.
I also enjoy helping others grow. A good code review doesn't just point out what's wrong, it leaves the other person understanding something they didn't before.
I care about understanding the problem, discussing decisions when they matter, and finding the simplest, most solid way to ship to production.
I work mainly with TypeScript and Node.js, and keep a close eye on AI and agents because the work is changing fast. I adopt new tools when they genuinely improve the product, the team, or how we work.
I'm finishing a Systems Engineering degree, though my career didn't wait for the diploma. Most of what I know I learned by building and getting things wrong on real systems.
Open to new projects, technical challenges and collaborations where the problem is real, whether you are a recruiter, founder or fellow engineer. Write through the form or email the address on this page. I work from Buenos Aires (GMT-3) and reply in English or Spanish.